Deep Learning for Segmentation of Cracks in High-Resolution Images of Steel Bridges

TL;DR

This paper introduces a deep learning approach combining ConvNext with an encoder-decoder network for crack segmentation in high-resolution steel bridge images, including a new dataset and a loss function to reduce false positives.

Contribution

The study presents a novel deep learning framework with a new dataset and a specialized loss function for improved crack detection in steel bridge images.

Findings

ConvNext enhances crack segmentation accuracy.

Background patches improve network performance.

New loss function reduces false positives.

Abstract

Automating the current bridge visual inspection practices using drones and image processing techniques is a prominent way to make these inspections more effective, robust, and less expensive. In this paper, we investigate the development of a novel deep-learning method for the detection of fatigue cracks in high-resolution images of steel bridges. First, we present a novel and challenging dataset comprising of images of cracks in steel bridges. Secondly, we integrate the ConvNext neural network with a previous state-of-the-art encoder-decoder network for crack segmentation. We study and report, the effects of the use of background patches on the network performance when applied to high-resolution images of cracks in steel bridges. Finally, we introduce a loss function that allows the use of more background patches for the training process, which yields a significant reduction in false positive rates.